Automated Stock Trading using Reinforcement Learning

Trading stocks requires effective methodologies to succeed in the business process. Automated stock trading involves utilizing computer algorithms and software programs to execute stock market trades automatically. This research work proposes an ensemble approach that leverages deep reinforcement learning to discover a stock trading strategy aimed at maximizing investment returns. By employing three actor-critic-based algorithms—Proximal Policy Optimization (PPO), Advantage Actor-Critic (A2C), and Deep Deterministic Policy Gradient (DDPG)—along with Soft actor-critic (SAC) and Twin Delayed DDPG (TD3), This work trains a deep reinforcement learning agent to develop an ensemble trading strategy. This ensemble strategy is designed to robustly adapt to various market conditions, as it inherits and integrates the most favorable characteristics from these three algorithms. Load on demand technique is supported in the proposed work in order to process extensive data efficiently and mitigate high memory usage during continuous action space network training, further this research employs a load-on-demand technique. The deep reinforcement learning agent is trained to implement an ensemble trading strategy that skillfully combines the best features of the three algorithms, thereby enabling it to adapt effectively to diverse market conditions. The results demonstrate that the proposed deep ensemble strategy consistently outperforms the three individual algorithms and the two baseline methods in terms of risk-adjusted returns, as indicated by the Sharpe ratio.